Actually, if you have the data, estimating failure rates of reactors is fairly straightforward. Just use a Poisson distribution, the same as calculating MTBFs for other rare events.

It's the same way one calculates the probability of being struck by lightening, for example.